Comparative analysis of two-dimensional data-driven efficient frontier estimation algorithms

Abstract

In this paper we show how the mathematical apparatus developed originally in the field of econometrics and portfolio optimization can be utilized for purposes of conceptual design, requirements engineering and technology roadmapping. We compare popular frontier estimation models and propose an efficient and robust nonparametric estimation algorithm for two-dimensional frontier approximation. The proposed model allows to relax the convexity assumptions and thus enable estimating a broader range of possible technology frontier shapes compared to the state of the art. Using simulated datasets we show how the accuracy and the robustness of alternative methods such as Data Envelopment Analysis and nonparametric and parametric statistical models depend on the size of the dataset and on the shape of the frontier.